Mutations

Adding or removing data in Dgraph is called a mutation.

A mutation that adds triples is done with the set keyword.

{
  set {
    # triples in here
  }
}

Triples

The input language is triples in the W3C standard RDF N-Quad format.

Each triple has the form

<subject> <predicate> <object> .

Meaning that the graph node identified by subject is linked to object with directed edge predicate. Each triple ends with a period. The subject of a triple is always a node in the graph, while the object may be a node or a value (a literal).

For example, the triple

<0x01> <name> "Alice" .

Represents that graph node with ID 0x01 has a name with string value "Alice". While triple

<0x01> <friend> <0x02> .

Represents that graph node with ID 0x01 is linked with the friend edge to node 0x02.

Dgraph creates a unique 64 bit identifier for every blank node in the mutation - the node’s UID. A mutation can include a blank node as an identifier for the subject or object, or a known UID from a previous mutation.

Blank Nodes and UID

Blank nodes in mutations, written _:identifier, identify nodes within a mutation. Dgraph creates a UID identifying each blank node and returns the created UIDs as the mutation result. For example, mutation:

{
 set {
    _:class <student> _:x .
    _:class <student> _:y .
    _:class <name> "awesome class" .
    _:x <name> "Alice" .
    _:x <planet> "Mars" .
    _:x <friend> _:y .
    _:y <name> "Bob" .
 }
}

results in output (the actual UIDs will be different on any run of this mutation)

{
  "data": {
    "code": "Success",
    "message": "Done",
    "uids": {
      "class": "0x2712",
      "x": "0x2713",
      "y": "0x2714"
    }
  }
}

The graph has thus been updated as if it had stored the triples

<0x6bc818dc89e78754> <student> <0xc3bcc578868b719d> .
<0x6bc818dc89e78754> <student> <0xb294fb8464357b0a> .
<0x6bc818dc89e78754> <name> "awesome class" .
<0xc3bcc578868b719d> <name> "Alice" .
<0xc3bcc578868b719d> <planet> "Mars" .
<0xc3bcc578868b719d> <friend> <0xb294fb8464357b0a> .
<0xb294fb8464357b0a> <name> "Bob" .

The blank node labels _:class, _:x and _:y do not identify the nodes after the mutation, and can be safely reused to identify new nodes in later mutations.

A later mutation can update the data for existing UIDs. For example, the following to add a new student to the class.

{
 set {
    <0x6bc818dc89e78754> <student> _:x .
    _:x <name> "Chris" .
 }
}

A query can also directly use UID.

{
 class(func: uid(0x6bc818dc89e78754)) {
  name
  student {
   name
   planet
   friend {
    name
   }
  }
 }
}

External IDs

Dgraph’s input language, RDF, also supports triples of the form <a_fixed_identifier> <predicate> literal/node and variants on this, where the label a_fixed_identifier is intended as a unique identifier for a node. For example, mixing schema.org identifiers, the movie database identifiers and blank nodes:

_:userA <http://schema.org/type> <http://schema.org/Person> .
_:userA <http://schema.org/name> "FirstName LastName" .
<https://www.themoviedb.org/person/32-robin-wright> <http://schema.org/type> <http://schema.org/Person> .
<https://www.themoviedb.org/person/32-robin-wright> <http://schema.org/name> "Robin Wright" .

As Dgraph doesn’t natively support such external IDs as node identifiers. Instead, external IDs can be stored as properties of a node with an xid edge. For example, from the above, the predicate names are valid in Dgraph, but the node identified with <http://schema.org/Person> could be identified in Dgraph with a UID, say 0x123, and an edge

<0x123> <xid> "http://schema.org/Person" .

While Robin Wright might get UID 0x321 and triples

<0x321> <xid> "https://www.themoviedb.org/person/32-robin-wright" .
<0x321> <http://schema.org/type> <0x123> .
<0x321> <http://schema.org/name> "Robin Wright" .

An appropriate schema might be as follows.

xid: string @index(exact) .
<http://schema.org/type>: [uid] @reverse .

Query Example: All people.

{
  var(func: eq(xid, "http://schema.org/Person")) {
    allPeople as <~http://schema.org/type>
  }

  q(func: uid(allPeople)) {
    <http://schema.org/name>
  }
}

Query Example: Robin Wright by external ID.

{
  robin(func: eq(xid, "https://www.themoviedb.org/person/32-robin-wright")) {
    expand(_all_) { expand(_all_) }
  }
}

External IDs and Upsert Block

The Upsert Block makes managing external IDs easy.

Set the schema.

xid: string @index(exact) .
<http://schema.org/name>: string @index(exact) .
<http://schema.org/type>: [uid] @reverse .

Set the type first of all.

{
  set {
    _:blank <xid> "http://schema.org/Person" .
  }
}

Now you can create a new person and attach its type using the Upsert Block.

   upsert {
      query {
        var(func: eq(xid, "http://schema.org/Person")) {
          Type as uid
        }
        var(func: eq(<http://schema.org/name>, "Robin Wright")) {
          Person as uid
        }
      }
      mutation {
          set {
           uid(Person) <xid> "https://www.themoviedb.org/person/32-robin-wright" .
           uid(Person) <http://schema.org/type> uid(Type) .
           uid(Person) <http://schema.org/name> "Robin Wright" .
          }
      }
    }

You can also delete a person and detach the relation between Type and Person Node. It’s the same as above, but you use the keyword “delete” instead of “set”. “http://schema.org/Person” will remain but “Robin Wright” will be deleted.

   upsert {
      query {
        var(func: eq(xid, "http://schema.org/Person")) {
          Type as uid
        }
        var(func: eq(<http://schema.org/name>, "Robin Wright")) {
          Person as uid
        }
      }
      mutation {
          delete {
           uid(Person) <xid> "https://www.themoviedb.org/person/32-robin-wright" .
           uid(Person) <http://schema.org/type> uid(Type) .
           uid(Person) <http://schema.org/name> "Robin Wright" .
          }
      }
    }

Query by user.

{
  q(func: eq(<http://schema.org/name>, "Robin Wright")) {
    uid
    xid
    <http://schema.org/name>
    <http://schema.org/type> {
      uid
      xid
    }
  }
}

Language and RDF Types

RDF N-Quad allows specifying a language for string values and an RDF type. Languages are written using @lang. For example

<0x01> <name> "Adelaide"@en .
<0x01> <name> "Аделаида"@ru .
<0x01> <name> "Adélaïde"@fr .

See also how language strings are handled in queries.

RDF types are attached to literals with the standard ^^ separator. For example

<0x01> <age> "32"^^<xs:int> .
<0x01> <birthdate> "1985-06-08"^^<xs:dateTime> .

The supported RDF datatypes and the corresponding internal type in which the data is stored are as follows.

Storage Type	Dgraph type
<xs:string>	string
<xs:dateTime>	dateTime
<xs:date>	datetime
<xs:int>	int
<xs:boolean>	bool
<xs:double>	float
<xs:float>	float
<geo:geojson>	geo
<xs:password>	password
<http://www.w3.org/2001/XMLSchema#string>	string
<http://www.w3.org/2001/XMLSchema#dateTime>	dateTime
<http://www.w3.org/2001/XMLSchema#date>	dateTime
<http://www.w3.org/2001/XMLSchema#int>	int
<http://www.w3.org/2001/XMLSchema#boolean>	bool
<http://www.w3.org/2001/XMLSchema#double>	float
<http://www.w3.org/2001/XMLSchema#float>	float

See the section on RDF schema types to understand how RDF types affect mutations and storage.

Batch mutations

Each mutation may contain multiple RDF triples. For large data uploads many such mutations can be batched in parallel. The command dgraph live does just this; by default batching 1000 RDF lines into a query, while running 100 such queries in parallel.

dgraph live takes as input gzipped N-Quad files (that is triple lists without { set {) and batches mutations for all triples in the input. The tool has documentation of options.

dgraph live --help

See also Fast Data Loading.

Delete

A delete mutation, signified with the delete keyword, removes triples from the store.

For example, if the store contained

<0xf11168064b01135b> <name> "Lewis Carrol"
<0xf11168064b01135b> <died> "1998"

Then delete mutation

{
  delete {
     <0xf11168064b01135b> <died> "1998" .
  }
}

Deletes the erroneous data and removes it from indexes if present.

For a particular node N, all data for predicate P (and corresponding indexing) is removed with the pattern S P *.

{
  delete {
     <0xf11168064b01135b> <author.of> * .
  }
}

The pattern S * * deletes all known edges out of a node (the node itself may remain as the target of edges), any reverse edges corresponding to the removed edges and any indexing for the removed data. The predicates to delete are derived from the type information for that node (the value of the dgraph.type edges on that node and their corresponding definitions in the schema). If that information is missing, this operation will be a no-op.

{
  delete {
     <0xf11168064b01135b> * * .
  }
}

Deletion of non-list predicates

Deleting the value of a non-list predicate (i.e a 1-to-1 relation) can be done in two ways.

1. Using the star notation mentioned in the last section.
2. Setting the object to a specific value. If the value passed is not the current value, the mutation will succeed but will have no effect. If the value passed is the current value, the mutation will succeed and will delete the triple.

For language-tagged values, the following special syntax is supported:

{
	delete {
		<0x12345> <[email protected]> * .
	}
}

In this example, the value of name tagged with language tag es will be deleted. Other tagged values are left untouched.

Mutations using cURL

Mutations can be done over HTTP by making a POST request to an Alpha’s /mutate endpoint. On the command line this can be done with curl. To commit the mutation, pass the parameter commitNow=true in the URL.

To run a set mutation:

curl -H "Content-Type: application/rdf" -X POST localhost:8080/mutate?commitNow=true -d $'
{
  set {
    _:alice <name> "Alice" .
  }
}'

To run a delete mutation:

curl -H "Content-Type: application/rdf" -X POST localhost:8080/mutate?commitNow=true -d $'
{
  delete {
    # Example: Alice's UID is 0x56f33
    <0x56f33> <name> * .
  }
}'

To run an RDF mutation stored in a file, use curl’s --data-binary option so that, unlike the -d option, the data is not URL encoded.

curl -H "Content-Type: application/rdf" -X POST localhost:8080/mutate?commitNow=true --data-binary @mutation.txt

JSON Mutation Format

Mutations can also be specified using JSON objects. This can allow mutations to be expressed in a more natural way. It also eliminates the need for apps to have custom serialisation code, since most languages already have a JSON marshalling library.

When Dgraph receives a mutation as a JSON object, it first converts in into multiple RDFs that are then processed as normal.

Each JSON object represents a single node in the graph.

Setting literal values

When setting new values, the set_json field in the Mutation message should contain a JSON object.

Literal values can be set by adding a key/value to the JSON object. The key represents the predicate, and the value represents the object.

For example:

{
  "name": "diggy",
  "food": "pizza"
}

Will be converted into the RDFs:

_:blank-0 <name> "diggy" .
_:blank-0 <food> "pizza" .

The result of the mutation would also contain a map, which would have the uid assigned corresponding to the key blank-0. You could specify your own key like

{
  "uid": "_:diggy",
  "name": "diggy",
  "food": "pizza"
}

In this case, the assigned uids map would have a key called diggy with the value being the uid assigned to it.

Language support

An important difference between RDF and JSON mutations is in regards to specifying a string value’s language. In JSON, the language tag is appended to the edge name, not the value like in RDF.

For example, the JSON mutation

{
  "food": "taco",
  "[email protected]": "tastes good",
  "[email protected]": "sabe bien",
  "[email protected]": "c'est bon",
  "[email protected]": "è buono"
}

is equivalent to the following RDF:

_:blank-0 <food> "taco" .
_:blank-0 <rating> "tastes good"@en .
_:blank-0 <rating> "sabe bien"@es .
_:blank-0 <rating> "c'est bon"@fr .
_:blank-0 <rating> "è buono"@it .

Geolocation support

Support for geolocation data is available in JSON. Geo-location data is entered as a JSON object with keys “type” and “coordinates”. Keep in mind we only support indexing on the Point, Polygon, and MultiPolygon types, but we can store other types of geolocation data. Below is an example:

{
  "food": "taco",
  location: {
    "type": "Point",
    "coordinates": [1.0, 2.0]
  }
}

Referencing existing nodes

If a JSON object contains a field named "uid", then that field is interpreted as the UID of an existing node in the graph. This mechanism allows you to reference existing nodes.

For example:

{
  "uid": "0x467ba0",
  "food": "taco",
  "rating": "tastes good",
}

Will be converted into the RDFs:

<0x467ba0> <food> "taco" .
<0x467ba0> <rating> "tastes good" .

Edges between nodes

Edges between nodes are represented in a similar way to literal values, except that the object is a JSON object.

For example:

{
  "name": "Alice",
  "friend": {
    "name": "Betty"
  }
}

Will be converted into the RDFs:

_:blank-0 <name> "Alice" .
_:blank-0 <friend> _:blank-1 .
_:blank-1 <name> "Betty" .

The result of the mutation would contain the uids assigned to blank-0 and blank-1 nodes. If you wanted to return these uids under a different key, you could specify the uid field as a blank node.

{
  "uid": "_:alice",
  "name": "Alice",
  "friend": {
    "uid": "_:bob",
    "name": "Betty"
  }
}

Will be converted to:

_:alice <name> "Alice" .
_:alice <friend> _:bob .
_:bob <name> "Betty" .

Existing nodes can be referenced in the same way as when adding literal values. E.g. to link two existing nodes:

{
  "uid": "0x123",
  "link": {
    "uid": "0x456"
  }
}

Will be converted to:

<0x123> <link> <0x456> .

Deleting literal values

Deletion mutations can also be sent in JSON format. To send a delete mutation, use the delete_json field instead of the set_json field in the Mutation message.

When using delete mutations, an existing node always has to be referenced. So the "uid" field for each JSON object must be present. Predicates that should be deleted should be set to the JSON value null.

For example, to remove a food rating:

{
  "uid": "0x467ba0",
  "rating": null
}

Deleting edges

Deleting a single edge requires the same JSON object that would create that edge. E.g. to delete the predicate link from "0x123" to "0x456":

{
  "uid": "0x123",
  "link": {
    "uid": "0x456"
  }
}

All edges for a predicate emanating from a single node can be deleted at once (corresponding to deleting S P *):

{
  "uid": "0x123",
  "link": null
}

If no predicates are specified, then all of the node’s known outbound edges (to other nodes and to literal values) are deleted (corresponding to deleting S * *). The predicates to delete are derived using the type system. Refer to the RDF format documentation and the section on the type system for more information:

{
  "uid": "0x123"
}

Facets

Facets can be created by using the | character to separate the predicate and facet key in a JSON object field name. This is the same encoding schema used to show facets in query results. E.g.

{
  "name": "Carol",
  "name|initial": "C",
  "friend": {
    "name": "Daryl",
    "friend|close": "yes"
  }
}

Produces the following RDFs:

_:blank-0 <name> "Carol" (initial=C) .
_:blank-0 <friend> _:blank-1 (close=yes) .
_:blank-1 <name> "Daryl" .

Creating a list with JSON and interacting with

Schema:

testList: [string] .

{
  "testList": [
    "Grape",
    "Apple",
    "Strawberry",
    "Banana",
    "watermelon"
  ]
}

Let’s then remove “Apple” from this list (Remember, it’s case sensitive):

{
   "uid": "0xd", #UID of the list.
   "testList": "Apple"
}

Add another fruit:

{
   "uid": "0xd", #UID of the list.
   "testList": "Pineapple"
}

Specifying multiple operations

When specifying add or delete mutations, multiple nodes can be specified at the same time using JSON arrays.

For example, the following JSON object can be used to add two new nodes, each with a name:

[
  {
    "name": "Edward"
  },
  {
    "name": "Fredric"
  }
]

JSON Syntax using Raw HTTP or Ratel UI

This syntax can be used in the most current version of Ratel, in the dgraph-js-http client or even via cURL.

You can also download the Ratel UI for Linux, macOS, or Windows.

Mutate:

{
  "set": [
    {
      # One JSON obj in here
    },
    {
      # Another JSON obj in here for multiple operations
    }
  ]
}

Delete:

Deletion operations are the same as Deleting literal values and Deleting edges.

{
  "delete": [
    {
      # One JSON obj in here
    },
    {
      # Another JSON obj in here for multiple operations
    }
  ]
}

Using JSON operations via cURL

First you have to configure the HTTP header to specify content-type.

-H 'Content-Type: application/json'

curl -H "Content-Type: application/json" -X POST localhost:8080/mutate?commitNow=true -d  $'
    {
      "set": [
        {
          "name": "Alice"
        },
        {
          "name": "Bob"
        }
      ]
    }' | jq

To delete:

curl -H "Content-Type: application/json" -X POST localhost:8080/mutate?commitNow=true -d  $'
    {
      "delete": [
        {
          "uid": "0xa"
        }
      ]
    }' | jq

Mutation with a JSON file:

curl -H "Content-Type: application/json" -X POST localhost:8080/mutate?commitNow=true -d @data.json

where the contents of data.json looks like the following:

{
  "set": [
    {
      "name": "Alice"
    },
    {
      "name": "Bob"
    }
  ]
}

The JSON file must follow the same format for mutations over HTTP: a single JSON object with the "set" or "delete" key and an array of JSON objects for the mutation. If you already have a file with an array of data, you can use jq to transform your data to the proper format. For example, if your data.json file looks like this:

[
  {
    "name": "Alice"
  },
  {
    "name": "Bob"
  }
]

then you can transform your data to the proper format with the following jq command, where the . in the jq string represents the contents of data.json:

cat data.json | jq '{set: .}'

{
  "set": [
    {
      "name": "Alice"
    },
    {
      "name": "Bob"
    }
  ]
}

Upsert Block

The Upsert block allows performing queries and mutations in a single request. The Upsert block contains one query block and one mutation block. Variables defined in the query block can be used in the mutation block using the uid function.

In general, the structure of the Upsert block is as follows:

upsert {
  query <query block>
  [fragment <fragment block>]
  mutation <mutation block>
}

The Mutation block currently only allows the uid function, which allows extracting UIDs from variables defined in the query block. There are 3 possible outcomes based on the results of executing the query block:

• If the variable is empty i.e. no node matched the query, the uid function returns a new UID in case of a set operation and is thus treated similar to a blank node. On the other hand, for delete/del, it returns no UID, and thus the operation becomes a no-op and is silently ignored.
• If the variable stores exactly one UID, the uid function returns the uid stored in the variable.
• If the variable stores more than one UID, the mutation fails. We plan to support this use case in the future.

Example

Consider an example with following schema:

curl localhost:8080/alter -X POST -d $'
  name: string @index(term) .
  email: string @index(exact) @upsert .
  age: int @index(int) .
  friend: uid @reverse .
' | jq

Insert Use Case

Now, let’s say we want to create a new user with email, and name information. We also want to make sure that two users cannot have same email id in the database.

We can do this using the Upsert block as follows:

curl -H "Content-Type: application/rdf" -X POST localhost:8080/mutate?commitNow=true -d  $'
upsert {
  query {
    me(func: eq(email, "[email protected]")) {
      v as uid
    }
  }

  mutation {
    set {
      uid(v) <name> "first last" .
      uid(v) <email> "[email protected]" .
    }
  }
}
' | jq

Result:

{
  "data": {
    "code": "Success",
    "message": "Done",
    "uids": {
      "uid(v)": "0x2"
    }
  },
  "extensions": {
    "server_latency": {
      "parsing_ns": 71033,
      "processing_ns": 22994018
    },
    "txn": {
      "start_ts": 4,
      "commit_ts": 5,
      "preds": [
        "1-email",
        "1-name"
      ]
    }
  }
}

The upsert first checks whether a user with the email [email protected] exists. Because the database is currently empty, no such user exists. Hence, the variable v will be empty. In this case, the uid function returns a new UID for the variable v replacing with the new UID value wherever uid(v) is used.

Update Use Case

Now, we want to add the age information for the same user having email ID [email protected]. We can use the Upsert block to do the same as follows:

curl -H "Content-Type: application/rdf" -X POST localhost:8080/mutate?commitNow=true -d  $'
upsert {
  query {
    me(func: eq(email, "[email protected]")) {
      v as uid
    }
  }

  mutation {
    set {
      uid(v) <age> "28" .
    }
  }
}
' | jq

Result:

{
  "data": {
    "code": "Success",
    "message": "Done",
    "uids": {}
  },
  "extensions": {
    "server_latency": {
      "parsing_ns": 39017,
      "processing_ns": 21231954
    },
    "txn": {
      "start_ts": 7,
      "commit_ts": 8,
      "preds": [
        "1-age"
      ]
    }
  }
}

Here, the query block queries for a user with email as [email protected]. It stores the uid of the user in variable v. The mutation block then updates the age of the user. The uid function extracts the uid from the variable v.